scvi.module.VAE

scvi.module.VAE#

class scvi.module.VAE(n_input, n_batch=0, n_labels=0, n_hidden=128, n_latent=10, n_layers=1, n_continuous_cov=0, n_cats_per_cov=None, dropout_rate=0.1, dispersion='gene', log_variational=True, gene_likelihood='zinb', latent_distribution='normal', encode_covariates=False, deeply_inject_covariates=True, batch_representation='one-hot', use_batch_norm='both', use_layer_norm='none', use_size_factor_key=False, use_observed_lib_size=True, extra_payload_autotune=False, library_log_means=None, library_log_vars=None, var_activation=None, extra_encoder_kwargs=None, extra_decoder_kwargs=None, batch_embedding_kwargs=None)[source]#

Bases: EmbeddingModuleMixin, BaseMinifiedModeModuleClass

Variational auto-encoder [Lopez et al., 2018].

Parameters:

  • n_input (int) – Number of input features.

  • n_batch (int (default: 0)) – Number of batches. If 0, no batch correction is performed.

  • n_labels (int (default: 0)) – Number of labels.

  • n_hidden (int (default: 128)) – Number of nodes per hidden layer. Passed into Encoder and DecoderSCVI.

  • n_latent (int (default: 10)) – Dimensionality of the latent space.

  • n_layers (int (default: 1)) – Number of hidden layers. Passed into Encoder and DecoderSCVI.

  • n_continuous_cov (int (default: 0)) – Number of continuous covariates.

  • n_cats_per_cov (list[int] | None (default: None)) – A list of integers containing the number of categories for each categorical covariate.

  • dropout_rate (float (default: 0.1)) – Dropout rate. Passed into Encoder but not DecoderSCVI.

  • dispersion (Literal['gene', 'gene-batch', 'gene-label', 'gene-cell'] (default: 'gene')) –

    Flexibility of the dispersion parameter when gene_likelihood is either "nb" or "zinb". One of the following:

    • "gene": parameter is constant per gene across cells.

    • "gene-batch": parameter is constant per gene per batch.

    • "gene-label": parameter is constant per gene per label.

    • "gene-cell": parameter is constant per gene per cell.

  • log_variational (bool (default: True)) – If True, use log1p() on input data before encoding for numerical stability (not normalization).

  • gene_likelihood (Literal['zinb', 'nb', 'poisson'] (default: 'zinb')) –

    Distribution to use for reconstruction in the generative process. One of the following:

  • latent_distribution (Literal['normal', 'ln'] (default: 'normal')) –

    Distribution to use for the latent space. One of the following:

    • "normal": isotropic normal.

    • "ln": logistic normal with normal params N(0, 1).

  • encode_covariates (bool (default: False)) – If True, covariates are concatenated to gene expression prior to passing through the encoder(s). Else, only gene expression is used.

  • deeply_inject_covariates (bool (default: True)) – If True and n_layers > 1, covariates are concatenated to the outputs of hidden layers in the encoder(s) (if encoder_covariates is True) and the decoder prior to passing through the next layer.

  • batch_representation (Literal['one-hot', 'embedding'] (default: 'one-hot')) –

    EXPERIMENTAL Method for encoding batch information. One of the following:

    • "one-hot": represent batches with one-hot encodings.

    • "embedding": represent batches with continuously-valued embeddings using Embedding.

    Note that batch representations are only passed into the encoder(s) if encode_covariates is True.

  • use_batch_norm (Literal['encoder', 'decoder', 'none', 'both'] (default: 'both')) –

    Specifies where to use BatchNorm1d in the model. One of the following:

    • "none": don’t use batch norm in either encoder(s) or decoder.

    • "encoder": use batch norm only in the encoder(s).

    • "decoder": use batch norm only in the decoder.

    • "both": use batch norm in both encoder(s) and decoder.

    Note: if use_layer_norm is also specified, both will be applied (first BatchNorm1d, then LayerNorm).

  • use_layer_norm (Literal['encoder', 'decoder', 'none', 'both'] (default: 'none')) –

    Specifies where to use LayerNorm in the model. One of the following:

    • "none": don’t use layer norm in either encoder(s) or decoder.

    • "encoder": use layer norm only in the encoder(s).

    • "decoder": use layer norm only in the decoder.

    • "both": use layer norm in both encoder(s) and decoder.

    Note: if use_batch_norm is also specified, both will be applied (first BatchNorm1d, then LayerNorm).

  • use_size_factor_key (bool (default: False)) – If True, use the obs column as defined by the size_factor_key parameter in the model’s setup_anndata method as the scaling factor in the mean of the conditional distribution. Takes priority over use_observed_lib_size.

  • use_observed_lib_size (bool (default: True)) – If True, use the observed library size for RNA as the scaling factor in the mean of the conditional distribution.

  • extra_payload_autotune (bool (default: False)) – If True, will return extra matrices in the loss output to be used during autotune

  • library_log_means (ndarray | None (default: None)) – ndarray of shape (1, n_batch) of means of the log library sizes that parameterize the prior on library size if use_size_factor_key is False and use_observed_lib_size is False.

  • library_log_vars (ndarray | None (default: None)) – ndarray of shape (1, n_batch) of variances of the log library sizes that parameterize the prior on library size if use_size_factor_key is False and use_observed_lib_size is False.

  • var_activation (Callable[[Tensor], Tensor] (default: None)) – Callable used to ensure positivity of the variance of the variational distribution. Passed into Encoder. Defaults to exp().

  • extra_encoder_kwargs (dict | None (default: None)) – Additional keyword arguments passed into Encoder.

  • extra_decoder_kwargs (dict | None (default: None)) – Additional keyword arguments passed into DecoderSCVI.

  • batch_embedding_kwargs (dict | None (default: None)) – Keyword arguments passed into Embedding if batch_representation is set to "embedding".

Notes

Lifecycle: argument batch_representation is experimental in v1.2.

Attributes table#

training

Methods table#

generative(z, library, batch_index[, ...])

Run the generative process.

loss(tensors, inference_outputs, ...[, ...])

Compute the loss.

marginal_ll(tensors, n_mc_samples[, ...])

Compute the marginal log-likelihood of the data under the model.

sample(tensors[, n_samples, max_poisson_rate])

Generate predictive samples from the posterior predictive distribution.

Attributes#

VAE.training: bool#

Methods#

VAE.generative(z, library, batch_index, cont_covs=None, cat_covs=None, size_factor=None, y=None, transform_batch=None)[source]#

Run the generative process.

Return type:

dict[str, Distribution | None]

VAE.loss(tensors, inference_outputs, generative_outputs, kl_weight=1.0)[source]#

Compute the loss.

Return type:

LossOutput

VAE.marginal_ll(tensors, n_mc_samples, return_mean=False, n_mc_samples_per_pass=1)[source]#

Compute the marginal log-likelihood of the data under the model.

Parameters:

  • tensors (dict[str, Tensor]) – Dictionary of tensors passed into forward().

  • n_mc_samples (int) – Number of Monte Carlo samples to use for the estimation of the marginal log-likelihood.

  • return_mean (bool (default: False)) – Whether to return the mean of marginal likelihoods over cells.

  • n_mc_samples_per_pass (int (default: 1)) – Number of Monte Carlo samples to use per pass. This is useful to avoid memory issues.

VAE.sample(tensors, n_samples=1, max_poisson_rate=100000000.0)[source]#

Generate predictive samples from the posterior predictive distribution.

The posterior predictive distribution is denoted as \(p(\hat{x} \mid x)\), where \(x\) is the input data and \(\hat{x}\) is the sampled data.

We sample from this distribution by first sampling n_samples times from the posterior distribution \(q(z \mid x)\) for a given observation, and then sampling from the likelihood \(p(\hat{x} \mid z)\) for each of these.

Parameters:

  • tensors (dict[str, Tensor]) – Dictionary of tensors passed into forward().

  • n_samples (int (default: 1)) – Number of Monte Carlo samples to draw from the distribution for each observation.

  • max_poisson_rate (float (default: 100000000.0)) – The maximum value to which to clip the rate parameter of Poisson. Avoids numerical sampling issues when the parameter is very large due to the variance of the distribution.

Return type:

Tensor

Returns:

Tensor on CPU with shape (n_obs, n_vars) if n_samples == 1, else (n_obs, n_vars,).

Contents